Exploring How Warmth and Competence Shape Trust in Medical Voice Agents with Blind Users
Abstract
As large language model (LLM)-based voice agents increasingly enter high-risk domains such as medical recommendation, voice personality is becoming a key social cue for trust. Yet research has largely focused on sighted users, leaving how blind users interpret these cues unclear. To address this gap, we conducted a two-part exploratory study with blind participants. We first created four theory-inspired voice probes varying intended warmth and competence, and ran a participatory co-design workshop (N=6) supported by tactile toolkits. This approach enabled participants to directly experience, evaluate, and articulate their perceptions. Our findings show how the two features (acoustic and linguistic) relate to perceived warmth and competence, which jointly shape trust judgments. To further examine these insights in a more realistic interaction context, we conducted a follow-up within-subject study (N = 11). Across both studies, blind users consistently prioritized competence over warmth in medical decision-making. Together, these findings contribute design knowledge and methods for building trustworthy and accessible voice agents in risk-sensitive contexts.